Quebec researchers use AI to encrypt data

SHERBROOKE, Que. – A team of researchers from the Université de Sherbrooke (UdeS) and the École de technologie supérieure (ETS) has developed a solution that will make the transmission and archiving of medical information more secure. Artificial intelligence is used to analyze all types of medical data, make diagnoses and even monitor certain treatments.

It could also be used therefore to encrypt and decrypt confidential information, according to the team led by Professor Pierre-Marc Jodoin (pictured) of the UdeS Faculty of Computer Science.

“We asked ourselves where future attacks against computer systems could come from,” he explains in an interview with The Canadian Press. “And (one of our findings) is that when our data is stored in the cloud, it’s not protected.”

The team’s work is based on the research of a former doctoral student supervised by Professor Jodoin.

Essentially, it involves training two artificial intelligence systems, known as “neural networks,” to encrypt medical data, making it incomprehensible to a human being.

“We have a first network that will deconstruct the data to allow it to leave the secure confines of the hospital or clinic and be transferred to a server,” explains the researcher.

“Once it’s been processed, a second network will produce a report that will also be incomprehensible, and send it to the clinician. The clinician will then have access to the neural network that can make sense of this data and enable it to be interpreted by the caregiver.”

Essentially, artificial intelligence will create a secret code to scramble the information and make it unusable for anyone who doesn’t hold the key.

Recent advances in artificial intelligence have opened up a world of possibilities for image processing. The researchers began their tests with medical imagery displaying human brains.

Each brain is unique and has a particular signature, a bit like a fingerprint,” said Jodoin. That’s why we took this data.”

The team’s research was presented in June at a conference in Argentina, where it was received enthusiastically.

The team would now like to test the model on a larger scale to validate its effectiveness.

“We’ve had access to brain data from around 500 subjects,” says Jodoin. “To take this to another level, we’d need to test it on public data from several tens of thousands of patients, which isn’t easy, as there aren’t many databases of this scale.”

“We’d also like to test it on other [types of data], such as CT scans, X-rays, ultrasound imaging, ultrasound scans and medical imaging of other parts of the body,” lists the professor. We’re [confident] that the technology will work just as well.”

However, they’ll have to be patient before their solution will be applied in healthcare settings, either here or somewhere else in the world.

“We’re doing the research,” says Jodoin. “Would a private company one day be interested in taking the technology, [appropriating it] and then go through a long approval process to commercialize it? It could take a few more years.”